混合语音段特征双边式优选算法用于帕金森病分类研究

发布时间：2018-07-16 20:53

【摘要】：近年来,已有研究证明基于语音数据可实现帕金森病(PD)的诊断,但是目前相关研究主要集中在特征提取及分类器设计等方面,对于样本优选方面考虑不足。本课题组前期研究结果表明,样本优选可有效改进分类准确性,但是样本和语音的相关关系至今还未能深入研究。因此,本文提出了基于相关特征加权和多核学习算法,同时对语音段和特征进行优选,用于发现语音段和特征的协同效应,从而达到提升PD分类准确性的目的。实验结果表明,本文算法针对受试者的分类准确率达到了82.5%,较已有文献算法提高了30.5%。此外,本文算法还挖掘出了语音段和特征的协同效应,对语音标记物提取有一定参考价值。
[Abstract]:In recent years, it has been proved that speech data can be used to diagnose Parkinson's disease (PD). However, the current research focuses on feature extraction and classifier design. Our previous research results show that sample selection can effectively improve the accuracy of classification, but the correlation between sample and speech has not been deeply studied. Therefore, based on correlation feature weighted and multi-core learning algorithm, the speech segment and feature are selected at the same time, which can be used to find the synergetic effect of speech segment and feature, so as to improve the accuracy of PD classification. The experimental results show that the classification accuracy of the proposed algorithm is 82.5%, which is 30.5% higher than that of the existing algorithms. In addition, the synergetic effect between speech segment and feature is mined by this algorithm, which has some reference value for speech marker extraction.
【作者单位】：重庆广播电视大学;重庆大学通信工程学院;第三军医大学西南医院神经内科;重庆医科大学第一附属医院神经内科;
【基金】：国家自然科学基金资助项目(61771080,61108086,91438104,61571069,61501065) 重庆市社会事业与民生保障科技创新专项(cstc2016shmszx40002) 重庆市基础与前沿技术研究专项(cstc2016jcyj A0043,cstc2016jcyj A0064,cstc2016jcyj A0134) 重庆市教委科学技术研究项目(KJ1603805,KJ1403808,KJ1503901) 西南医院联合孵化项目(SWH2016LHYS-11) 中央高校基本科研业务费(10611CDJXZ238826)
【分类号】：R742.5;TP181

【相似文献】